Every infra-tool company is shipping AI right now. Most of them are shipping the same thing: an LLM that acts — restarts services, applies patches, kills processes — and asks for forgiveness later.
We don’t ship that. AI Doctor reads. It explains. It suggests. It never runs a destructive command on its own.
What AI Doctor actually does
When a deployment fails, AI Doctor:
- Reads the build log, the runtime log, the recent commits, and the service config.
- Compares the failure pattern against ones it’s seen succeed before.
- Names the most likely cause — in plain English, with the evidence.
- Suggests one or two specific fixes.
- Stops.
If you click “apply fix,” the change goes through the same approval + audit flow as if you’d typed it yourself. There is no path where AI Doctor side-effects on your behalf.
Why “assist, never act”
The autonomy question isn’t a capability question — current LLMs are absolutely capable of restarting nginx. The question is what happens when the model is wrong.
When a developer is wrong, they pause, ask, sanity-check. When an LLM is wrong, it acts with the same confidence as when it’s right. The asymmetric risk between “AI suggests a fix you reject” and “AI runs a fix that destroys data” is roughly infinite. We took the position that AI sits next to the operator, not in place of them.
Where AI does run free
Reading is fine. AI Doctor can scan thousands of log lines, correlate across hosts, summarize patterns — that’s pure observation. Same with threat-intel correlation, anomaly detection, trend prediction. The “act” boundary is where the line gets drawn, not the “think” one.
This is the third value on our list, with a sharper edge: AI is allowed to talk. It’s not allowed to break things. Stays that way.